Reproduce “Neuroevolution of a Recurrent Neural Network for Spatial and Working Memory in a Simulated Robotic Environment” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xinyun Zou, Eric O. Scott, Alexander B. Johnson, Kexin Chen, Douglas A. Nitz, Kenneth A. De Jong, Jeffrey L. Krichmar
Researchers successfully applied neuroevolution to develop a recurrent neural network controller that enables a simulated robot to solve a complex spatial and working memory task in a triple T-maze. The evolved network demonstrated efficient navigation, robust spatial encoding, and prospective trajectory coding, with performance primarily driven by its internal dynamics rather than individual sensory inputs.
Key insights
- Complex spatial and working memory capabilities can emerge from the neuroevolution of recurrent neural networks, demonstrating the power of adaptive systems for higher-order cognitive functions.
- The internal dynamics of the evolved RNNs, integrating information across time and modalities, are the primary drivers of successful navigation and memory, rather than reliance on individual sensory inputs.
- The population activity of the RNN neurons creates a robust spatial representation (cognitive map) and encodes prospective information about future actions, paralleling biological neural codes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.